As AI agents have recently evolved from simple tools to independent platform participants, a major headache has emerged: Sybil attacks, where an infinite number of fake accounts are created to disrupt ecosystems. If a malicious user were to unleash 10,000 agents to manipulate ratings, skew community votes, and seize service traffic, how could we stop them? Traditional CAPTCHAs are simply no match for such overwhelming automated attacks.

Fortunately, global standards organizations and developer communities are moving quite quickly to address this. The W3C has already launched a group to discuss a dedicated DID registry for agents, and the OpenID Foundation has published a draft of AIIM, an agent token standard. Recently, an international standard led by South Korea and the U.S., with the FBI participating as a co-editor, was even adopted. The goal is to provide agents with cryptographically verified identities and permissions, clearly establishing which agents are real and who owns or controls them.

{
  "iss": "https://identity.agent.org",
  "sub": "did:key:z6Mkg...",
  "agent_owner": "did:ion:EiD3...",
  "delegated_permissions": ["read:profile", "write:feed"],
  "trust_registry": "https://moltrust.base"
}

Example based on OpenID Foundation AIIM Draft 00

However, verifying identity alone does not guarantee trustworthiness. An interesting approach here is a reputation system like AgentRank. Instead of simple star ratings that anyone can manipulate, it calculates a reputation graph based on the settlement value of micro-payments or real financial transactions that an agent has actually processed. In short, it mathematically verifies whether the agent is a real partner that has made genuine transactions and paid its dues. By combining this with human-challenge oracle techniques—which leverage finite human cognitive resources to significantly increase the cost of automation—a powerful line of defense is being established.

As Google's agent-to-agent communication protocol, A2A, gains attention, developers are deeply considering how to build their own trust verification layers. In the era of the agent economy, where agents will act as full-fledged members of platforms, these cryptographic identity and transaction-based reputation systems seem to be an essential survival strategy rather than just an option.